Hacker News Reader

Comment by seunosewa | Hacker News Reader

Today I learned that Google's brain teasers are internally known as "estimation or market sizing questions"!

Asking "how would you estimate the number of cars sold in a year?" is not a brainteaser. It's an estimation / market sizing question. You can deduce a reasonable estimate from the population size and other assumptions.

A brainteaser is something like "A man pushed his car to a hotel and lost his fortune. What happened?" That's a brainteaser -- and nothing Google would ask.

He's playing Monopoly!

So, do I get the job?

Is there anyone that have read one of the readily available interview books that can't answer one of those estimation questions? I'm sure the answer to that question is 'yes', but I am also sure that their utter inability to prepare for an interview or reason on their feet will be discovered in the interview by other means.

"How would you estimate the number of cars sold in a year?"

This boggles my mind. Where? For what purpose? Is the expectation that the candidate googles that question?

I would google lots of different things until I found some answers (car manufacturers sales reports from car magazines, news articles that may have the answer, literally for the phrase "number of cars sold in a year" etc, etc).

If you want me to talk through it, it's just mind numbing making asinine assumptions. Market sizing? Obvious assumptions? What?

I can't understand how anyone can say with a straight face that that question is straight forward and not ambiguous.

The ambiguity is what makes these google interview questions brain teasers, to me at least.

But that's probably the last item in a list of reasons why Google wouldn't hire me :)

It's okay that it's ambiguous. You can ask questions, like whether or not they're including motorcycles and buses.

The point is not to get the right answer. It's to deduce a reasonable answer that's in the right ballpark.

Consider the Google engineer who implemented the OneBox feature that answers "how many cars are sold in a year?"

How did he do it?

Consider the Wolfram Alpha engieer who implemented the same feature.

How did she do it?

(It becomes an interesting question, eh?)

But they didn't? Google doesnt answer it remotely close to what I find by searching deeper with each.

It's so ambiguous at the google search engine can't answer it.

The purpose is to see how you break down the problem. It's all about applying MECE (Mutually Exclusive, Collectively Exhaustive) methodology.

Usually for this type of questions you don't really even need to say the numbers, just clearly outline your though process.

The ambiguity is what makes it relevant to search.

What does a user mean if they enter [how many cars sold in a year?]

The ambiguity shows that the specific answer is less interesting than showing your working; having the discussion about how to get an answer.

That's more what I was hoping for, thanks :)

I wonder what internal Google has explored along the lines of prompting the user with clarifying questions. Like Google Suggest and Spell Check exhibit the level of intelligence needed I think.

Ambiguity is something a programmer needs to deal with at work. Riddles are not. Open ended questions are exactly what you want to use to evaluate someone. Riddles fail because there is an expected answer, not because the form of the question is ambiguous.

Some coding questions as asked in interviews are actually riddles in practice because the interviewer is unskilled. Those 'coding questions' are worse than the obviously bad riddles because it is harder to tell that the feedback is slanted.

I got asked the cars sold in a year in California question. I answered by saying I would look for a car that was new two years ago, and one that was new this year, and then difference the license plate numbers and divide by 2.

Took the interviewer a while but they finally figured it out.

What do you mean by 'difference the license plate numbers'?

Chuck is assuming (reasonably) that license plate numbers are assigned in sequential order to all new cars.

So if 1000 cars are sold in a year, maybe he'll see plate #267 from Year 0, and plate #1796 from Year 2. 1800 - 300 / 2 = 750 cars per year, close enough.

Exactly, in California private plates are of the form /[0-9][A-Z]{3}[0-9]{3}/ you can think of it as three digits base 10, and three digits base 26, one digit base 10. There are letter combos they don't use like SUX or FUX etc but each of those only reduce the pool by 1000 vehicles for each banned letter sequence so overall a relatively small hit on the estimate. My wife pointed out you can be more accurate if you include the month of the renewal in the registration (its on the plate as well) to get total months in your sample.

Reply on news.ycombinator.com